What’s Vulnerable › Vulnerabilities

CVE-2026-115913: vLLM: Phi-4 multimodal audio encoder materializes and retains full attention matrices: a single max-length audio OOM-crashes the vLLM engine

• Severity: medium (CVSS 6.5 v3.1)
• Exploited in the wild: not listed in CISA KEV
• EPSS: 0.3% probability of exploitation in the next 30 days
• Public exploit or detection template: yes
• Published: 2026-10-26 · updated 2026-10-28
• Weakness: CWE-770

Description

vLLM is an inference and serving engine for large language models. Prior to 0.32.0, the Phi-4 multimodal audio encoder in phi4mm_audio.py and phi4mm_utils.py uses eager attention by default, allowing a remote client to submit one legal long audio clip that causes MultiHeadedAttention.forward_attention to materialize and retain full attention matrices in self.attn. The scaled dot-product attention branch does not materialize these matrices. The resulting quadratic GPU-memory use can cause an out-of-memory condition and terminate EngineCore for all users. This issue is fixed in version 0.32.0.

Affected products

• vllm-project vllm: from 0.31.0 before 0.32.0 — fixed in 0.32.0
• vllm-project vllm: before 0.32.0

How to fix

Upgrade to a fixed version: vllm-project vllm 0.32.0. Check the vendor advisory for mitigations.

References

• https://github.com/vllm-project/vllm/commit/3d52578e13454de0bb8352cac988347cdcc89928
• https://github.com/vllm-project/vllm/pull/60614
• https://github.com/vllm-project/vllm/releases/tag/v0.32.0
• https://github.com/vllm-project/vllm/security/advisories/GHSA-8397-8qf6-f2gm

Get alerted when CVE-2026-115913 or the next vulnerability affects your stack